Skip to content

Latest commit

 

History

History
67 lines (55 loc) · 3.83 KB

how_to_use_longiseg.md

File metadata and controls

67 lines (55 loc) · 3.83 KB

How to use LongiSeg

LongiSeg inherits nnU-Net’s self-configuring capabilities, allowing it to easily adapt to new datasets. It extends nnU-Net to handle medical image time series, leveraging temporal relationships to improve segmentation accuracy.

Dataset Format

LongiSeg expects the same structured dataset format as nnU-Net. Datasets can either be saved in the same folders as nnU-Net or in LongiSeg's own folder (LongiSeg_raw, LongiSeg_preprocessed and LongiSeg_results). In contrast to nnU-Net, LongiSeg expects an additional patientsTr.json file in the dataset folder. This file lists patient IDs and their corresponding scans in chronological order.

Dataset001_BrainTumour/
├── dataset.json
├── patientsTr.json
├── imagesTr
├── imagesTs  # optional
└── labelsTr

This json file should have the following structure:

{
    "patient_1": [
        "patient_1_scan_1",
        "patient_1_scan_2",
        ...
    ],
    "patient_2": [
        "patient_2_scan_1",
        "patient_2_scan_2",
        ...
    ],
    ...
}

Experiment planning and preprocessing

To run experiment planning and preprocessing of a new dataset, simply run

LongiSeg_plan_and_preprocess -d DATASET_ID

or if you prefer to keep things separate, you can also use LongiSeg_extract_fingerprint, LongiSeg_plan_experiment and LongiSeg_preprocess (in that order). We refer to the nnU-Net documentation for additional details on experiment planning and preprocessing.

Training

To train a model using LongiSeg, simply run

LongiSeg_train DATASET_NAME_OR_ID UNET_CONFIGURATION FOLD

By default, LongiSeg uses the LongiSegTrainer, which integrates the temporal dimension by concatenating multi-timepoint images as additional input channels. Other trainers are available with the -tr option:

  • nnUNetTrainerNoLongi: A modified nnUNetTrainer where training data is split at the patient level instead of per scan. (non-longitudinal baseline)
  • LongiSegTrainerDiffWeighting: A longitudinal trainer that incorporates the Difference Weighting Block for temporal feature fusion.
  • LongiSegTrainerRP, LongiSegTrainerDiffWeightingRP: longitudinal trainer with randomly sampled instead of fixed prior scan of the same patient (c.f. longi_dataset)

Other options for training are available as well (LongiSeg_train -h).

Inference

Inference with LongiSeg works in a similar way to the nnU-Net inference, with the added requirement of specifying a patient file (-pat) in either the LongiSeg_predict or LongiSeg_predict_from_modelfolder commands. The patient file needs to detail the patient structure in the same way as during training. Only cases present in both the input folder and patients.json will be processed during inference!

LongiSeg_predict -i INPUT_FOLDER -o OUTPUT_FOLDER -path /path/to/patients.json -d DATASET_ID

Evaluation

By default LongiSeg performs nnU-Net's standard evaluation on the 5-fold cross validation. This, however, does not account for individual patients and only calculates a handful of metrics. LongiSeg extends nnU-Net’s evaluation by incorporating additional metrics that provide a more comprehensive assessment of segmentation performance, including volumetric, surface-based, distance-based, and detection metrics. To run evaluation with LongiSeg, use

LongiSeg_evaluate_folder GT_FOLDER PRED_FOLDER -djfile /path/to/dataset.json -pfile /path/to/plans.json -patfile /path/to/patients.json

If no patient file is provided, LongiSeg will default to standard nnU-Net evaluation while incorporating the additional metrics.